Characterizing the time course of automatic action tendencies to affective facial expressions and its dysregulation in social anxiety disorder

Abstract

Affective facial expressions elicit automatic approach or avoidance action tendencies, which are dysregulated in Social Anxiety Disorder (SAD). However, research has not dissociated the initiation and execution phases of automatic action tendencies, which may be distinctly modulated by affective faces and SAD. In Study 1, fifty adults completed a modified Approach-Avoidance Task (AAT) that characterized the time course of automatic approach or avoidance actions elicited by affective faces. In the initiation phase, happy faces elicited greater automatic approach tendencies compared to angry faces, an effect that linearly weakened across the execution phase. In Study 2, 44 adults with a principal diagnosis of SAD and 22 healthy comparison (HC) adults completed a similar AAT. Compared to the HC group, the SAD group exhibited an inconsistent time course of automatic action tendencies to neutral faces. Specifically, SAD was characterized by relatively weak initiation of automatic approach tendencies, but relatively stronger execution of automatic approach tendencies. In contrast, the HC group exhibited relatively similar initiation and execution of automatic approach tendencies to neutral faces. Together, these results demonstrate that the initiation and execution of automatic action tendencies are differentially modulated by affective faces and SAD.

Introduction

Automatic action tendencies putatively support the initiation of overt social behaviors, which may be dysregulated in SAD (Loijen, Vrijsen, Egger, Becker, & Rinck, 2020; Strack & Deutsch, 2004). Automatic action tendencies operate largely outside of conscious awareness to rapidly decrease (automatic approach) or increase (automatic avoidance) interpersonal distance to affective stimuli such as facial expressions (Krieglmeyer et al., 2013). Broadly, automatic approach actions facilitate overt approach behaviors, whereas automatic avoidance actions facilitate overt avoidance behaviors (Kawakami, Phills, Steele, & Dovidio, 2007). In non-anxious individuals, happy facial expressions elicit automatic approach tendencies, whereas angry facial expressions elicit automatic avoidance tendencies (Stins et al., 2011). In Social Anxiety Disorder (SAD), however, individuals exhibit dysregulated automatic action tendencies, which can disrupt overt social behaviors independent of self-reported perception (Heuer, Rinck, & Becker, 2007; Roelofs et al., 2010; Taylor & Amir, 2012). For example, individuals with SAD exhibit slower automatic approach actions to neutral facial expressions compared to non-anxious individuals (Kuckertz, Strege, & Amir, 2017). Furthermore, dysregulated automatic action tendencies are associated with dysregulated social behavior at sub-clinical levels of social anxiety (Taylor & Amir, 2012). Together, these findings suggest that dysregulated automatic action tendencies in SAD may facilitate maladaptive social behaviors. As such, issues related to the assessment of automatic action tendencies may offer important clinical implications for understanding the role of automatic action tendencies in SAD.

Experimental paradigms typically measure automatic action tendencies by manipulating the perceived distance between an individual and an affective stimulus such as a facial expression. By manipulating interpersonal distance, paradigms simulate approach towards or avoidance away from a stimulus. In joystick paradigms, such as the Approach-Avoidance Task (AAT), participants push or pull a joystick to increase or decrease the visual size of a face (Rinck & Becker, 2007). By increasing or decreasing the size of facial expressions, these paradigms aim to simulate visual changes associated with approaching or avoiding stimuli. In manikin paradigms, participants press buttons that increase or decrease the visual distance between an individual’s virtual manikin avatar and a stimulus (e.g., Krieglmeyer & Deutsch, 2010). In short, paradigms manipulate interpersonal distance from either a first-person perspective (joystick paradigms) or third-person perspective (manikin paradigms).

When approach or avoidance actions are explicitly directed by an emotion relevant response contingency (e.g., happy = approach; angry = avoid), happy faces reliably elicit faster approach actions, whereas angry faces reliably elicit faster avoidance actions (Phaf, Mohr, Rotteveel, & Wicherts, 2014). However, this response contingency directly links actions to identification of facial emotion, which precludes disentangling automatic action tendencies from perception of stimuli. To ensure that automatic action tendencies are implicitly influenced by affective properties of stimuli, participants are instructed to approach or avoid stimuli independent of valence. For example, participants may approach or avoid affective facial expressions based on stimulus background color (e.g., green = approach; blue = avoid). Although these response contingencies ensure that automatic action tendencies are implicitly modulated, happy and angry facial expressions do not reliably elicit differential automatic action tendencies in these paradigms (Phaf et al., 2014). Inconsistent results challenge the conceptualization that affective facial expressions automatically elicit differential actions, which complicates understanding the role that automatic actions play in SAD.

To clarify inconsistent findings, it may be important to consider the time course of automatic action tendencies. The first stage of a motor response represents the latency required to cognitively process a stimulus (e.g., a colored background), select a response (e.g., pushing or pulling a joystick), and initiate the selected response (e.g., start pushing or pulling). The second stage of a motor response represents the latency between the initiation of the selected motor response and the subsequent execution of this motor response to completion (e.g., continue pushing a joystick until trial concludes), which is also referred to as “movement time” or “movement execution” (Lavender & Hommel, 2007). In short, the initiation stage of a motor response comprises both stimulus processing and response selection, whereas the execution stage involves comprises continued execution of the selected response. Given the differences in cognitive demand between these stages, initiation of an action is characterized by a notably longer latency compared to the subsequent execution of an action (Coombes, Cauraugh, & Janelle, 2007). Thus, automatic action tendencies unfold along a time course that is characterized by differential involvement of cognitive processes.

Research demonstrates that affective stimuli differentially influence the early initiation of a motor action, but do not modulate the subsequent execution of a motor action (for a review, see Beatty, Cranley, Carnaby, & Janelle, 2016). However, most paradigms produce reaction time (RT) measures that combine the initiation and execution of automatic action tendencies. For example, joystick paradigms typically assess RTs after participants fully complete an extended pull or push movement (i.e., from stimulus onset to offset). As a result, joystick paradigms measure the combined initiation and execution of automatic action tendencies. In contrast, manikin paradigms primarily measure the RT required to initiate avatar movement, whereas the subsequent execution RTs required to complete the movement are not analyzed (Krieglmeyer & Deutsch, 2010). When comparing these paradigms directly, affective stimuli more reliably modulate initiation RTs in the manikin paradigm compared to combined RTs in the joystick paradigm (Krieglmeyer & Deutsch, 2010). Therefore, affective stimuli may modulate the initiation, but not subsequent execution, of automatic action tendencies. Given that these studies did not directly compare initiation and execution measures, however, the time course of automatic action tendencies remains uncharacterized.

Additionally, individual difference factors also distinctly modulate the initiation and execution of automatic action tendencies, which offers important implications for SAD. For example, trait anger is associated with faster initiation of automatic approach actions to angry facial expressions, but not the subsequent execution of actions (Veenstra, Schneider, Bushman, & Koole, 2017). Similarly, social avoidance behavior modulates the initiation, but not subsequent execution of automatic actions (Evans & Britton, 2020). Thus, SAD may also distinctly modulate the initiation and execution of automatic action tendencies. For example, individuals with SAD may struggle to engage in overt approach behaviors specifically due to impaired initiation of automatic approach actions to facial expressions. Given the recent development of promising treatment protocols that aim to modify dysregulated automatic action tendencies in SAD (Taylor & Amir, 2012), it is important to characterize the time course of SAD-related differences.

However, no research has directly characterized the time course of automatic action tendencies to affective facial expressions or SAD-related differences. To address these issues, we conducted two independent studies. In Study 1, we developed a modified AAT paradigm in which automatic action tendencies are measured using multiple, discrete responses to iteratively approach or avoid affective facial expressions. By using successive discrete responses, it is possible to characterize the time course of automatic action tendencies from initiation throughout subsequent execution. First, we hypothesized that happy faces would initially elicit greater automatic approach relative to angry and neutral faces, whereas angry faces would initially elicit greater automatic avoidance relative to happy and neutral faces. Second, we hypothesized that emotion-related differences in automatic actions would linearly weaken across the execution of subsequent motor responses. In Study 2, we used a modified joystick AAT paradigm to characterize SAD-related differences in the initiation and execution of automatic actions to affective facial expressions. Based on previous AAT research in SAD (Kuckertz et al., 2017), we hypothesized that SAD would be characterized by slower initiation of automatic approach tendencies to neutral facial expressions compared to a healthy control (HC) group, but similar execution of automatic approach tendencies.

Study 1

Method

Participants

Fifty-one adults (Age = 21.20 years, SD = 6.45, Range = 18 – 52; 32 females) were recruited from the University of Miami and surrounding community. We selected our sample size based on meta-analytic effect sizes of 0.15 reported in the AAT literature (Laham, Kashima, Dix, & Wheeler, 2015). Participants were 18 years or older in age, English-speaking, and reported normal or corrected-normal vision. Normative color vision was confirmed using the Ishihara Test of Color Deficiency (Ishihara, 1917).

Procedure

As part of a larger study, participants completed a modified version of the AAT, which was implemented using Visual Basic Advance programming within E-Prime 2.0 software (Psychology Software Tools, Inc.). All study procedures and compensation were approved and conducted in compliance with local Institutional Review Board (IRB) guidelines, including informed consent.

Measures

Approach-Avoidance Task (AAT)

In this paradigm, affective facial expressions were presented against a colored background (see Figure 1A). Happy, angry, and neutral faces were selected from 12 individuals (6 males, 6 females) in the NimStim Face Stimulus set (Tottenham et al., 2009). During each trial, facial expressions were presented on either a blue or green background. Thus, each trial served as either an “approach” trial or an “avoid” trial, which was communicated via background color. Based on this background color, participants made successive key presses that either increased (approach) or decreased (avoidance) the size of facial expressions to simulate approaching or avoiding stimuli (Krieglmeyer & Deutsch, 2010).

Figure 1. Sample Trials of Modified Approach-Avoidance Task (AAT) Paradigms.

Participants were presented with facial expressions that appeared on a blue or green background. In Study 1 (A), each button press increased/decreased the size of the image by 20%. Reaction time (RT) collected from response 1 measures the initiation of automatic actions, whereas RTs collected from responses 2–5 measure the subsequent execution of automatic actions. In Study 2 (B), participants were instructed to respond via joystick based on the background color of the facial expression. RT collected from the initial movement of the joystick beyond the central fixation box measures the initiation of automatic actions, whereas RT collected beyond the fixation box to trial completion measures the subsequent execution of automatic actions.

Each trial began with a medium-sized facial expression. Participants were instructed to press a mouse key with an index finger (e.g., blue = left, green = right) until the image disappeared. With each response, the image either increased or decreased in size by 20% increments. Participants were provided with a 2000ms response window to produce five consecutive responses. Response latency and accuracy were separately recorded for each response. A blank screen was presented between trials for a jittered inter-trial interval of 500ms (Range: 250ms – 750ms).

Each facial expression was presented as both an approach and an avoidance trial. Color assignment (e.g., blue = approach, green = avoid) and button assignment (e.g. left = approach; right = avoid) was counterbalanced across participants. Across two blocks, participants completed 144 trials (72 Approach and 72 Avoid). Before the task, participants were provided with instructions and 20 practice trials with neutral facial expressions (1 male, 1 female), which required at least 70% accuracy before continuing the study.

Data Reduction

Incorrect responses and reaction times (RTs) greater than 2.5 standard deviations from a participant’s mean Approach or Avoid RT were excluded. Additionally, one participant was excluded due to obtaining an accuracy rate lower than 70%. Across the final sample, data cleaning procedures removed 6.4% of trials.

For each facial expression, separate approach-avoidance scores were computed for each response within a trial, which were subsequently averaged across trials (i.e., RT_{ApproachHappy1} – RT_AvoidHappy1, RT_{ApproachHappy2} – RT_AvoidHappy2, etc.). Positive scores indicate an automatic approach tendency, whereas negative scores indicate an automatic avoidance tendency.

Data Analysis Strategy

Time Course of Automatic Action Tendencies

We hypothesized that affective facial expressions would elicit emotion-related differences in initial approach-avoidance scores (e.g., Happy₁ > Angry₁) that would decrease linearly across execution of responses (e.g., [Happy₁ vs. Angry₁] > [Happy₂ vs. Angry₂] > … > [Happy₅ vs. Angry₅]). To characterize the time course of automatic action tendencies, approach-avoidance scores were submitted to a 3 (Emotion: Happy, Angry, and Neutral) × 5 (Response: 1–5) omnibus Repeated Measures ANOVA (RM-ANOVA). Specifically, we compared linear changes in emotion-related differences across responses using a 3 (Emotion: Happy > Neutral > Angry) × Linear (Response: 1 > 2 > 3 > 4 > 5) polynomial contrast. Following significant interactions (a ≤ 0.05), polynomial contrasts were used to compare linear changes in automatic action tendencies between each pair of emotions (e.g., Happy vs. Angry).

Post-Hoc Comparison of Automatic Action Tendencies

After confirming the dissociation between initiation and execution in our paradigm (see Supplemental Material), we utilized paired-samples t-tests (one-sided) to confirm that affective facial expressions elicited differences in initial approach-avoidance scores (i.e., Response #1), but not subsequent approach-avoidance scores (i.e., Responses #2-#5). Specifically, we hypothesized that happy facial expressions would initially elicit greater automatic approach tendencies relative to both angry and neutral facial expressions. We hypothesized that angry facial expressions would initially elicit greater automatic avoidance tendencies relative to both happy and neutral facial expressions. In contrast, we hypothesized that the execution of automatic actions would not be modulated by affective facial expressions.

Results

Time Course of Automatic Action Tendencies

To test for emotion-related differences in the time course of automatic action tendencies, we first analyzed automatic action tendencies across all 5 responses. As hypothesized, a significant Emotion (Happy > Neutral > Angry) × linear Response (1 > 2 > 3 > 4 > 5) interaction emerged (F_1,49 = 5.14; p = 0.03; η_p² = 0.10; see Figure 2). No emotion-related interactions were observed with non-linear Response trends (both p’s > 0.13).

Figure 2. Time Course of Automatic Action Tendencies.

Automatic approach-avoidance scores were computed by indexing mean reaction times on Avoidance trials from mean reaction times (RTs) on Approach trials (i.e., RT_Avoid – RT_Approach). For each trial, an approach-avoidance score was computed for each of the 5 responses. Positive scores indicate an automatic approach tendency, whereas negative scores indicate an automatic avoidance tendency. At the first response, approach-avoidance scores differed across emotions, which linearly weakened across subsequent responses (p = 0.03). For descriptive purposes, results are also depicted in the form of mean RTs separately for Approach and Avoidance trials.

To decompose the significant omnibus Emotion × Linear Response interaction, we compared the linear Response trends between each emotion pair (e.g., Linear Happy vs. Linear Angry; see Figure 3). For the Happy-Angry contrast (HA), an Emotion × Linear Response interaction indicated that happy faces tended to elicit greater initial automatic approach tendencies compared to angry faces, which weakened across responses (HA₁ > HA₂ > HA₃ > HA₄ > HA₅; F_1,49 = 3.97; p = 0.05; η_p² = 0.08). For the Happy-Neutral contrast (HN), a non-significant Emotion × Linear Response interaction indicated that automatic approach tendencies elicited by happy faces also tended to weaken across responses (HN₁ > HN₂ > HN₃ > HN₄ > HN₅; F_1,49 = 3.48; p = 0.07; η_p² = 0.07). For the Angry-Neutral contrast (AN), however, angry faces did not elicit initial automatic avoidance tendencies compared to neutral faces, which remained consistent across responses (F_1,49 = 0.07; p = 0.79; η_p² = 0.001).

Figure 3. Time Course of Decreasing Emotion-Related Differences in Automatic Action Tendencies.

Emotional contrasts were computed by comparing approach-avoidance scores between emotions at each response (e.g., Happy₁ – Angry₁, Happy₂ – Angry₂, etc.). Dashed lines depict the linear trend for each emotion contrast. Automatic approach tendencies initially elicited by happy facial expressions weakened across responses compared to angry (p = 0.05) and neutral facial expressions (p = 0.07). Automatic avoidance tendencies elicited by angry facial expressions did not weaken across responses compared to neutral facial expressions (p = 0.79).

Post-Hoc Comparison of Automatic Action Tendencies

Initial response

Paired-samples t-tests (one-sided) were conducted to compare initial approach-avoidance scores (i.e., RT1_Approach – RT1_Avoid) between each emotion. Happy faces (M = 9.63, SD = 47.98) elicited significantly greater approach tendencies relative to angry faces (M = −3.00, SD = 42.67; t(49) = 1.72, p < 0.05; see Figure 2) Although in the expected direction, happy faces did not elicit significantly greater approach tendencies relative to neutral faces (M = −1.54, SD = 49.52; t(49) = 1.36, p = 0.09). Finally, angry faces did not elicit significantly different automatic action tendencies relative to neutral faces (t(49) = −0.19, p = 0.43).

Subsequent execution responses

Paired-samples t-tests (one-tailed) were also conducted to compare subsequent approach-avoidance scores (e.g., Avg[Happy_2–5] vs. Avg[Angry_2–5]). Averaged across subsequent responses, happy faces (M = 12.46, SD = 16.86) did not elicit approach tendencies that differed relative to angry faces (M = 14.16, SD = 18.13; t(49) = −1.27, p = 0.11). Moreover, this pattern of results was in the opposite direction compared to the results observed within initial responses (i.e., Subsequent: Angry > Happy; Initiation: Happy > Angry). Similarly, happy faces (M = 12.46, SD = 16.86) did not elicit approach tendencies that differed relative to neutral faces (M = 13.57, SD = 20.23) across subsequent responses (t(49) = −0.81, p = 0.21). Averaged across subsequent responses, angry faces did not elicit greater automatic avoidance tendencies relative to neutral faces (t(42) = 0.42, p = 0.34). Notably, exploratory analyses demonstrated descriptive emotion-related differences within the execution phase (see Supplemental Material).

Study 1 Summary

In Study 1, affective facial expressions evoked differences in the initiation, but not the subsequent execution, of automatic action tendencies. Specifically, happy faces initially elicited greater automatic approach action tendencies relative to angry faces, but this difference linearly weakened across the subsequent execution of actions. Although a similar pattern of results was observed when happy facial expressions were compared to neutral facial expressions, the effects were not statistically significant. In contrast, angry faces did not elicit greater automatic avoidance action tendencies relative to neutral faces either initially or across the subsequent execution of actions.

Study 2

Method

Participants

As part of a larger treatment protocol, this study recruited a total of 66 socially anxious and non-anxious adults from the San Diego community. Sample size was chosen based on a priori power calculations from the larger protocol to detect differences in automatic action tendencies between the HC and SAD groups with a large effect size (d = 0.78). Inclusion criteria for the SAD group were a current principal diagnosis of SAD as determined by both a diagnostic interview and a score ≥ 50 on the clinician administered Liebowitz Social Anxiety scale (LSAS). The Social Anxiety Disorder group (SAD) was comprised of 44 treatment-seeking adults (Age = 23.05 years, SD = 4.65; 65.9% women; 2.30% Non-Binary; 43.20% White, 34.10% Asian, 6.80% multi-racial, 6.80% Black/African American, 4.50% unknown or declined, 2.30% American Indian/Alaska Native, 2.30% “other”). The healthy control group (HC) included 22 adults (Age = 26.86 years, SD = 5.78, 68.2% women; 57.10% White, 28.60% Asian, 14.30% multi-racial) with no current or past psychiatric diagnoses and a score of ≤ 20 on the LSAS. For both groups, exclusionary criteria included: active suicidality, moderate/severe alcohol or marijuana use disorder (past year), mild other substance use disorders (past year), bipolar disorder, psychotic disorder, traumatic brain injury, impairing medical conditions, lacking English proficiency, psychotropic medications, concurrent psychotherapy (unless 12-week stability criteria met for non-empirically supported therapies), or MRI contraindications.

Procedure

All participants completed an AAT paradigm during MRI scanning. Notably, fMRI data are being prepared for submission as a separate manuscript and are outside the scope of the current study. Importantly, all data reported in the current study were collected prior to treatment assignment. All study procedures, informed consent, and compensation were approved and conducted in compliance with local Institutional Review Board (IRB) guidelines.

Measures

Diagnostic Assessments

To establish diagnostic criteria, participants were administered the SAD module of the Structured Clinical Interview for DSM-IV (SCID-IV; First, Spitzer, Gibbon, & Williams, 2002). To confirm SAD as the principal diagnosis and assess co-morbid diagnoses, participants were also administered the Mini International Neuropsychiatric Interview (MINI; Sheehan et al., 1998). All diagnostic assessments were administered by a PhD-level clinician, doctoral level trainee, or post-baccalaureate clinical research coordinator. Diagnostic consensus was determined by reviewing interviews with the second author.

Liebowitz Social Anxiety Scale

To dimensionally assess social anxiety symptoms, participants were administered the clinician-rated LSAS (Liebowitz & Pharmacopsychiatry, 1987). The LSAS is a 24-item scale consisting of various social situations (e.g., talking on the phone in public). Participants rate the degree to which they experience fear responses (0 = None; 3 = Severe) and avoid (0 = Never; 3 = Always) these situations. These two scales are summed to measure social anxiety severity, which exhibited excellent internal consistency in the current sample (Cronbach’s alpha = .98).

Beck Depression Inventory-II

To assess depressive symptoms, participants completed the self-report version of the BDI-II (Beck, Steer, & Brown, 1996). The BDI-II assesses the degree to which individuals experienced depressive symptoms over the previous two weeks, which exhibited excellent internal consistency in the current sample (Cronbach’s alpha = .95).

AAT Paradigm

The AAT paradigms in Study 1 and Study 2 shared several core features that facilitate indirect comparison of results (see Figure 1B). First, this AAT paradigm presented happy, angry, and neutral facial expressions. Second, each AAT trial began with a medium-sized facial expression framed by a blue or green border. Third, participants were instructed to increase (approach) or decrease (avoid) the size of facial expressions based on the border color.

Nevertheless, it is also important to note that the AAT paradigm in Study 2 also differed from Study 1 in several aspects. First, facial expressions were comprised of 8 actors selected from the Karolinska Directed Emotional Faces stimulus set (KDEF; Goeleven, De Raedt, Leyman, & Verschuere, 2008). Second, approach/avoidance actions were made by pulling or pushing a joystick controller, rather than a computer mouse. Third, participants did not make successive, discrete responses, but instead completed each trial by pushing or pulling a joystick controller that fluidly increased or decreased image size. Specifically, Study 1 utilized 5 successive, discrete responses on each trial using a computer mouse, whereas Study 2 used a continuous push/pull motion on each trial using a joystick controller (see details below). Finally, the AAT paradigms presented different numbers of trials (Study 1: 144 trials; Study 2: 192 trials).

Prior to each trial, participants were instructed to center the joystick controller in the upright position. To facilitate centering the joystick, a small square was presented on the screen and a fixation cross indicated the joystick’s relative position. Participants were instructed to center the joystick by moving the fixation cross within this square (see Figure 1B). Upon stimulus presentation, participants pulled or pushed the joystick until the image disappeared, which occurred when the joystick moved 30° in either direction.

Data Reduction

Data reduction steps for Study 2 were identical to those utilized in Study 1. No participants were excluded due to poor accuracy (all accuracy rates > 82.81%). However, one participant (HC control) was excluded due to failing to properly center the joystick before at least 70% of trials. Across the final sample, data cleaning procedures removed 8.07% of trials.

To parallel Study 1, RTs were dissociated into initiation and execution responses. Initiation RTs were measured by the latency between stimulus onset and participant’s initial response to push or pull the joystick controller. Specifically, initiation RTs were recorded when the joystick position was moved outside the central fixation square (see Figure 1B). Therefore, initiation RTs collected in Study 2 are analogous to RTs of the first response collected in Study 1. Execution RTs were measured by the latency between the joystick position moving outside the central fixation square and the final 30° position. Therefore, execution RTs in Study 2 are analogous to the average RTs of responses 2–5 in Study 1. We computed separate approach-avoidance scores for initiation RTs (e.g., RT_{ApproachHappy1} – RT_AvoidHappy1) and execution RTs (e.g., RT_{ApproachHappy2} – RT_AvoidHappy2).

Data Analytic Strategy

Sample Characteristics

We examined SAD-related differences in demographic variables (age, gender, ethnicity, and race), social anxiety symptoms (LSAS), and depressive symptoms (BDI-II). Given the small and unequal frequencies of racial identities between groups, we tested racial identitiy into as a dichotomous variable.

SAD-Related Differences in Automatic Action Tendencies

In Study 2, we aimed to characterize SAD-related differences in the time course of automatic action tendencies to affective facial expressions. SAD-related differences in automatic action tendencies are observed in response to neutral facial expressions (Kuckertz et al., 2017). However, sub-clinical levels of social anxiety are associated with automatic avoidance tendencies in response to happy facial expressions and/or angry facial expressions (Heuer et al., 2007; Lange, Keijsers, Becker, & Rinck, 2008). Based on these heterogeneous findings, we compared SAD-related differences in the time course of automatic action tendencies to each emotional facial expression separately.

To characterize SAD-related differences in the time course of automatic action tendencies, automatic action tendency scores for each facial expression (Angry, Happy, Neutral) were submitted to separate 2 (Group: SAD vs. HC) × 2 (Response: Initiation vs. Execution) Repeated Measures ANOVAs. Following significant SAD-related interactions (a = 0.05), we also examined the time course within each group.

Using this analytic strategy, SAD-related differences in automatic action tendencies are not directly compared between emotions. To account for individual differences in general RT and its time course, we computed general indices of initiation RT (Mean[Happy_Initiation, Angry_Initiation, Neutral_Initiation]), execution RT (Mean[Happy_Execution, Angry_Execution, Neutral_Execution]), and time course of RT across phases (RT_Initiation - RT_Execution). We repeated each Phase × Group model while controlling for these general RT measures.

Results

Sample Characteristics

Demographics.

The SAD group and HC group did not differ in ethnicity, racial composition, or gender distribution (all p’s > 0.30). However, the SAD group (M = 23.05, SD = 4.64) was unexpectedly younger in age compared to the HC group (M = 27.10, SD = 1.27; t(63) = 3.03, p = 0.004). However, all results remained identical when controlling for age.

Symptom Measures.

The SAD group (M = 84.27, SD = 16.99) demonstrated significantly higher social anxiety symptoms compared to the HC group (M = 7.48, SD = 5.47; t(57.84) = 26.72, p < 0.001). Additionally, the SAD group (M = 18.45, SD = 11.55) demonstrated significantly higher depressive symptoms compared to the HC group (M = 1.57, SD = 1.69; t(45.58) = 9.48, p < 0.001).

Given SAD-related differences in depressive symptoms, it is possible that depressive symptoms contribute to SAD-related differences, which is difficult to disentangle (Miller & Chapman, 2001). Nevertheless, we conducted post-hoc analyses to address this issue. First, we repeated the Phase × Group interaction model on neutral facial expressions while including BDI-II as a continuous covariate. Second, we repeated the Phase × Group interaction model on neutral facial expressions while excluding SAD participants who met criteria for a current co-morbid diagnosis of Major Depressive Disorder (MDD; n = 12).

SAD-Related Differences in Automatic Action Tendencies

Similar to Study 1, all 3 emotion models exhibited a significant main effect of Phase in which automatic action tendencies were significantly more skewed towards approach during the execution phase compared to the initiation phase (Happy: [F(_{1, 63}) = 33.26, p < 0.001], Angry: [F(_{1, 63}) = 62.80, p < 0.001], Neutral: [F(_{1, 63}) = 17.88, p < 0.001]; see Figure 4). Across all 3 emotion models, we did not observe any significant main effects of SAD group (Happy: [F(_{1, 63}) = 1.47, p = 0.24], Angry: [F(_{1, 63}) = 0.47, p = 0.50], or Neutral: [F(_{1, 63}) = 0.06, p = 0.81]).

Figure 4. Social Anxiety-Related Differences in the Time Course of Automatic Action Tendencies.

Automatic approach-avoidance scores were computed by indexing mean reaction times on Avoidance trials from mean reaction times (RTs) on Approach trials (i.e., RT_Avoid – RT_Approach). For each trial, an approach-avoidance score was computed separately for the initiation and execution phases. Positive scores indicate an automatic approach tendency, whereas negative scores indicate an automatic avoidance tendency. In response to neutral faces, socially anxious individuals exhibited weaker automatic approach tendencies during the initiation phase, but stronger automatic approach tendencies during the execution phase (p = 0.009).

For neutral facial expressions, we observed the hypothesized significant Phase × Group interaction effect (F(_{1, 63}) = 7.30, p = 0.009; see Figure 4). Importantly, the Phase × Group interaction remained significant within control analyses that covaried for general RT indices (F(_{1, 61}) = 4.01, p = 0.050). To decompose this interaction, we examined between-group differences across the time course of automatic action tendencies to neutral facial expressions. During the initiation phase, the SAD group (M = 20.10; SD = 50.04) exhibited weaker automatic approach tendencies compared to the HC group (M = 42.014, SD = 52.99). During the execution phase, however, the SAD group (M = 83.73; SD = 70.08) exhibited stronger automatic approach tendencies compared to the HC group (M = 56.02; SD = 37.06).

To further characterize the Phase × Group interaction, we also examined the time course of automatic action tendencies to neutral facial expressions within each group. Within the HC group, automatic approach tendencies did not vary in strength across the initiation phase and execution phase (F(_{1, 20}) = 1.32, p = 0.26). Within the SAD group, however, automatic approach actions significantly strengthened across the initiation and execution phases (F(_{1, 43}) = 31.96, p < 0.001; see Figure 4).

For happy and angry facial expressions, we did not observe a significant Phase × Group interaction (Happy: [F(_{1, 63}) = 3.53, p = 0.07], Angry: [F(_{1, 63}) = 2.54, p = 0.12]). Moreover, the Phase × Group interactions were significantly attenuated within RT control analyses (Happy: [F(_{1, 61}) = 1.20, p = 0.28], Angry: [F(_{1, 61}) = 0.49, p = 0.49]). Given that the Phase × Group interaction for happy facial expressions was marginally significant and did not survive control analyses, we did not probe this interaction further.

Post-Hoc Analyses

Controlling for Depressive Symptomatology

When simultaneously including continuous BDI-II scores and SAD diagnosis in the Neutral facial expression model, we did not observe a significant BDI-II × Phase interaction (F(_{1, 61}) = 0.29, p = 0.60). Similarly, we did not observe a significant main effect of BDI-II scores (F(_{1, 61}) = 0.28, p = 0.60). However, controlling for continuous BDI-II scores attenuated the Phase × Group interaction (F(_{1, 61}) = 3.04, p = 0.09).

When excluding SAD participants diagnosed with a current MDD, the Phase × Group interaction remained significant (p = 0.02). Moreover, models including current MDD as a dummy-coded covariate continued to demonstrate significant Phase × Group interaction (p = 0.02), but not a MDD × Phase interaction (p = 0.85).

Study 2 Summary

In Study 2, we observed SAD-related differences in the time course of automatic action tendencies specifically to neutral facial expressions, which is consistent with previous research (Kuckertz et al., 2017). During the initial phase, the SAD group was characterized by weaker automatic approach tendencies compared to the HC group. During the execution phase, however, the SAD group was characterized by stronger automatic approach tendencies compared to the HC group. Within-group analyses demonstrated that SAD was characterized by automatic approach tendencies that significantly increased in strength across the initiation and execution phases. In contrast, the HC group demonstrated a more consistent time course of automatic approach tendencies, which did not significantly differ between the initiation and execution phases.

General Discussion

Across two studies, the initiation and execution of automatic action tendencies were differentially influenced by affective facial expressions (Study 1) and SAD (Study 2). In Study 1, affective facial expressions elicited differential automatic action tendencies during the initiation phase, which subsequently weakened across the execution phase. In Study 2, neutral facial expressions elicited opposite patterns of SAD-related differences across the initiation and execution phase. Compared to the HC group, SAD was characterized by weaker initiation of automatic approach tendencies to neutral facial expressions, but stronger execution of automatic approach tendencies. Together, these results demonstrate the importance of dissociating the time course of automatic action tendencies.

The initiation and execution phases of automatic action tendencies may represent differential contributions of multiple processes. During the initiation phase, the prepotent response elicited by a stimulus (e.g., happy face = approach) may be either congruent (e.g., green background = approach) or incongruent (e.g., blue background = avoid) with action selection. When prepotent responses must be inhibited to guide action selection, affective facial expressions may exert a relatively large influence on the latency to initiate automatic action tendencies (Kaldewaij, Koch, Volman, Toni, & Roelofs, 2016). Following action selection, however, subsequent execution responses do not require selection between competing actions. Instead, subsequent execution of actions only requires sustained performance of a basic motor response (e.g., button press). Across this subsequent execution phase, action selection may be increasingly dominated by the mental representation of the previous motor response (e.g., left button press). Due to the increasing influence of previous response on action selection during the execution phase, prepotent responses elicited by stimuli may exert a decreasing influence on subsequent execution of actions. More complex techniques that employ drift diffusion models may be helpful to more precisely characterize the relative contributions of these processes across the time course of automatic action tendencies (Krypotos, Beckers, Kindt, & Wagenmakers, 2015).

In both studies, approach responses were increasingly faster than avoidance responses across the time course of automatic action tendencies. Although not hypothesized, research has generally demonstrated that approach responses are faster than avoidance responses independent of stimulus valence (Heuer et al., 2007; Krieglmeyer & Deutsch, 2013; Paulus & Wentura, 2016). Notably, approach responses are generally faster than avoidance responses across different types of stimuli (e.g., facial expressions and objects) as well as different response methods (e.g., joystick and virtual manikin). In our studies, increasingly faster approach responses may be attributable to the visual manipulation of stimulus size, which differs between approach and avoid trials. Specifically, stimuli and background colors increased in size with each approach response, but decreased in size with each avoidance response. Thus, the primary response contingency (i.e., background color) may become increasingly salient with each approach response relative to each avoidance response, which increasingly skews automatic action tendencies.

In Study 2, we observed a complex pattern of SAD-related differences in response to neutral facial expressions that diverged across the initiation and execution phases. Based on previous research (Kuckertz et al., 2017), we originally hypothesized that SAD-related differences in automatic action tendencies would be observed to neutral facial expressions during the initiation phase, but not the subsequent execution phase. Instead, we observed that SAD-related differences were observed in opposite directions across the initiation and execution phases. Within the initiation phase, individuals with SAD demonstrated an initial hesitation to approach neutral facial expressions as evidenced by weaker automatic approach tendencies compared to the HC group. Within the execution phase, however, individuals with SAD demonstrated stronger automatic approach tendencies compared to the HC group. In contrast, non-anxious individuals in the HC group exhibited a more consistent time course of automatic approach tendencies to neutral facial expressions. Specifically, the HC group was characterized similar automatic approach tendencies across the initiation and execution responses. Thus, SAD may be characterized by an inconsistent time course of automatic action to neutral facial expressions in which hesitant initiation of automatic approach tendencies is followed an accelerated execution of automatic approach tendencies.

In SAD, an inconsistent time course of automatic approach tendencies could be interpreted in two, functionally distinct manners. Following an initial hesitation to approach neutral facial expressions, individuals with SAD may accelerate automatic approach tendencies across the execution phase to counteract the initial hesitation. In this view, accelerated execution of automatic approach tendencies in SAD may compensate for an initial hesitation to approach neutral faces, which adaptively adjusts automatic actions. In the AAT, however, completing the execution of a motor response removes the facial expression from view. Throughout the execution phase, neutral facial expressions increase in size to simulate approach towards the facial expression. As a result, accelerated execution of automatic approach tendencies may alternatively function to reduce aversive exposure to neutral faces in SAD. In short, accelerated execution to automatically approach neutral facial expressions may serve an adaptive or maladaptive function in SAD. To disentangle these possibilities, future work could utilize AAT paradigms that do not remove images from view, which would rule out the interpretation that acceleration of automatic approach actions serves to reduce aversive exposure (Bartoszek & Winer, 2015).

Although the current study cannot disentangle these functional interpretations, the inconsistent time course of automatic approach tendencies observed in SAD may nevertheless offer important clinical implications. First, implicit treatment protocols have been developed that aim to modify dysregulated automatic action tendencies to normalize social behavior. At sub-clinical levels of social anxiety, these implicit treatment protocols implicitly train individuals to automatically approach happy facial expressions, which subsequently increases engagement during social interactions (Taylor & Amir, 2012). At clinical levels of SAD, however, research suggests that implicit protocols fail to normalize social behaviors (Asnaani, Rinck, Becker, & Hofmann, 2014). One potential explanation for equivocal clinical outcomes is that individuals with SAD may exhibit a failure to automatically approach ambiguous neutral facial expressions, rather than a failure to automatically approach unambiguous happy facial expressions (Kuckertz et al., 2017). Consistent with this conceptualization, our results suggest that individuals with SAD may exhibit hesitant initiation, but accelerated execution, of automatic approach actions to neutral facial expressions. In conjunction with previous research, our findings suggest that individuals with SAD may benefit from implicit treatment protocols that aim to improve the balance between initiation and execution of automatic approach actions to neutral facial expressions.

Although we believe these findings offer important clinical implications, several limitations should be noted. First, we did not observe automatic avoidance in response to angry facial expressions in Study 1, which is consistent with past work that recruited unselected samples (Seidel, Habel, Kirschner, Gur, & Derntl, 2010). However, Study 2 recruited a sample with SAD, which similarly did not demonstrate automatic avoidance to angry facial expressions. One potential explanation for such asymmetrical findings is that angry facial expressions may elicit automatic avoidance or automatic approach actions depending on individual differences or contextual factors (Carver & Harmon-Jones, 2009; Krieglmeyer & Deutsch, 2013; Veenstra et al., 2017). As such, we can not conclude that automatic avoidance tendencies weaken throughout the execution of automatic actions. Second, although the AAT paradigms used in Study 1 and Study 2 were similar in many important characteristics, these paradigms were nevertheless not identical. Thus, similarities in results between studies should be considered descriptive pending direct replication.

Despite these limitations, the current study offers important implications for characterizing both automatic action tendencies and the role of these automatic actions in SAD. Characterizing the time course of automatic actions revealed that affective facial expressions differentially influence the initiation and execution of action. Moreover, this approach revealed that SAD was most accurately characterized by an inconsistent time course of automatic action tendencies to neutral facial expressions. Ultimately, dissociating the time course of automatic action tendencies may offer important insights into both adaptive and maladaptive social behavior.

Highlights.

• Affective faces elicited automatic action tendencies in two distinct phases

• Happy and angry faces elicited differential automatic actions in the initiation phase

• Differential automatic actions linearly weakened across the execution phase

• SAD was characterized by an inconsistent time course of action to neutral faces

• SAD was characterized by weaker initiation and stronger execution of approach